Netflix has detailed how it is building a reliability platform for large-scale GPU training and inference, using layered health checks, automated remediation, and fault-tolerant training to keep jobs running despite hardware failures. By Craig Risi